NVIDIA RTX A6000 & RTX A40 48GB GDDR6 Graphics Cards Announced


Nvidia Quadro RTX A6000 and RTX A40 with a GA102 GPU, up to 40 TFlops and 48GB GDDR6, should provide a lot of steam in the data center and workstation.

Nvidia recently announced RTX A6000 and the RTX A40 graphics card based on the Ampere architecture. Both the video cards use the full configuration of the GA102 graphics chip, which already drives the GeForce RTX 3080 and RTX 3090. The two professional cards remain in the two-slot format and with a Thermal Design Power (TDP) of 300 watts despite the higher computing power than the RTX 3090.

The Nvidia RTX A6000 is a classic, almost 27 centimeter long 2-slot graphics card with a radial fan that blows the heated air out of the case (direct exhaust of heat with blower). The A40 does not have a fan, but is supposed to use the usually strong airflow in server cases.

Both video cards use the full configuration of the GA102 graphics chip with 84 shader multiprocessors and 10,752 cores – 2 SMs and 256 cores more than the top gaming model, the GeForce RTX 3090. Their TDP is sealed off at 300 watts, which is why they also have a single 8-Pin power connector.

Nvidia didn’t specify any clock rates for the graphics chip in advance, but the possible computing power can be judged. This is usually calculated on the basis of the boost clock, which, however, may not be able to maintained consistently with a tight TDP budget. The figures for the rounded 40 and 38 TFlops should therefore be viewed as rather optimistic, but are still 10% and 5% respectively above the GeForce RTX 3090, which rarely has to throttle below its 1.7 GHz boost with 350 watts.

Both the NVIDIA RTX A6000 & RTX A40 have 48GB memory and they are provided with ECC memory protection. In addition, it is “only” the conventional GDDR6 type, not the new GDDR6X. With GDDR6X, 48GB would currently not be possible due to the lack of 2 GB memory chips. Nvidia equips both cards with the full 384 data lines of the integrated memory controller, but clocks the RAM a little lower on the A40. Two cards at a time can be connected with NVLink (aggregated 112.5 GByte/s). Please see the table below for full specifications.


Both cards should also be usable for virtualization – including with Nvidia Grid, the Quadro Virtual Data Center Workstation and the Virtual Compute Server. This enables virtual GPUs with 1, 2, 3, 4, 5, 6, 12, 16, 24, and 48GB to be created. The virtualization only works when the display ports are switched off. On the A40 they are switched off at the factory setting and must first be activated in the software, on the Nvidia RTX A6000 it is the other way around.

Like the Founders Edition of the GeForce RTX 3000, the RTX A6000 also does without the USB-C connection Virtual Link. It connects monitors up to 8K resolution via four DisplayPorts 1.4. At the request of customers who would like to do without the actively ventilated Quadros in data centers, the A40 can be equipped with display outputs. However, as a passively cooled card, it only has three DP 1.4a, so that it can make better use of the airflow in the server housing.

The NVIDIA RTX A6000 should be available from December, while the first OEM systems with the RTX A40 should go on sale in early 2021. Nvidia did not mention prices in advance, but the Quadro RTX 8000 with 48GB and Turing graphics chip currently costs around $5,181 on Amazon.

Nvidia graphics cards for workstations and data centers
RTX A6000 A40 RTX 8000 GV100
Cuda-cores 10752 10752 4608 5120
Tensor cores 336 336 576 640
Boost clock ? ? 1770MHz ~1450MHz
Memory speed 16Gbps GDDR6 14.5Gbps GDDR6 14Gbps GDDR6 1,7Gbps HBM2
Memory bus 384-bit 384-bit 384-bit 4096-bit
Vram 48GB 48GB 48GB 32GB
Ecc Partly (dram) Partly (dram) Partly (dram) Fully
Half Precision ? ? 32.6 TFLOPS 29.6 TFLOPS
Single-Precision ? ? 16.3 TFLOPS 14.8 TFLOPS
Tensor Performance ? ? 130.5 TFLOPS 118.5 TFLOPs
TDP 300W 300W 295W 250W
Cooling Active Passive Active Passive
NVLink 1x NVLink3
1x NVLink3
1x NVLInk2
2x NVLInk2
GPU GA102 GA102 TU102 GV100
Architecture Ampere Ampere Turing Volta
Manufacturing Process Samsung 8nm Samsung 8nm TSMC 12nm FFN TSMC 12nm FFN
Launch Price ? ? $10,000 $9,000
Launch Date 12/2020 Q1 2021 Q4 2018 March 2018


Shibu is a contributor here especially for gaming topics and has the technical expertise also to write on mobiles, technology, hardware, etc.